From APIs to AI: How MCP Changes the Way Systems Integrate with Business Central 

The “Golden Rule” of systems integration has been simple: if you want to talk to Business Central (BC), you use an API. Whether it was OData, SOAP, or even the newer REST endpoints, developers write code to integrate between external systems and the ERP. And it worked. But it was rigid, expensive, and hard to maintain. 

But now, we find ourselves at the precipice of an enormous change. With the advent of Generative AI and LLMs, it is no longer sufficient to simply expose an API. We need systems that can understand context, not just exchange data. 

Enter The Model Context Protocol (MCP). 

In this article, we’re going to explore how MCP is changing the game in terms of Business Central integration. 

The Old Way: The “API Spaghetti” Problem 

To understand why MCP is a gamechanger, we first need to understand the problems with the status quo. 

• Creating a custom API extension in AL code. 

• Securing the endpoint via OAuth/Azure Entra ID. 

• Creating a new “Tool” or “Function” in your AI code to call that specific API. 

• Hardcoding the logic to tell the AI when to call that API. 

The pain point here is that it’s a lot of work. Every time you want to do something new like posting a journal entry or checking the customer’s inventory, you have to do it all over again.  

Enter MCP: The “USB Port” for AI 

MCP can be thought of as a new “standard,” similar in concept to USB and Bluetooth, but for the data systems and AI. 

Rather than requiring glue code for every single interaction between the AI and the data systems, the MCP provides a standardized and open-source protocol that allows the AI assistant to query the data systems securely. 

In the context of Business Central, the MCP is a dynamic translation layer that allows the AI model to query your ERP in real-time, without requiring you to create a specific API handler. 

How MCP Alters Business Central Integrations 

So, what are the implications for the BC developer or consultant? 

1. “From Hardcoded Endpoints to Dynamic Discovery” 

With traditional APIs, the AI is only able to do what you program into it. But what if you have an MCP server connected into Business Central? The AI can now “discover” what’s available. 

For example, if the MCP server exposes the Business Central “Customer” table, the AI can automatically determine how to query the No., Name, or Balance fields without you having to write code like get_customer_balance. 

2. Context Aware Interactions 

Standard APIs are stateless; they don’t have any knowledge of the conversation history. 

MCP is built with the goal of being context aware. 

Scenario: “Who owes us the most money?” 

Traditional: The API may simply return a list of customers. 

MCP: The AI can use the protocol to first query the Detailed Cust. Ledg. Entry table, compute the open balances in real-time, and then ask the user, “Do you want me to send a reminder email to the top 3 overdue accounts?” The integration is not just a simple retrieval; the API is participating in the process. 

3. Secure, Governed Access 

One of the biggest fears of integrating AI with ERP is security. You do not want a chat session with an AI to accidentally change your General Ledger setup. 

MCP servers operate locally or in your infrastructure. This means you can use standard BC security permissions. If the user of the AI question being asked does not have permission to DELETE in the Sales Header table, the MCP server will simply not allow it. They bring their standard BC permission set into the AI conversation. 

4. Eliminating “Connector Fatigue” 

Today, to integrate a niche app into BC, we must develop a connector. With MCP, however, if a third-party app (such as a niche app for inventory scanning or a niche app for human resources) supports the MCP standard, it will instantly connect to your Business Central AI environment. 

The Future is “Agentic” 

MCP is not just integration; it’s “Agency.” 

Business Central was once just a database, just waiting for the application code to tell it what to do. Now, with the protocol, Business Central is an active participant in the world of business intelligence. An AI agent can always monitor your data and only alert you, when necessary, like when the stock level of an item plummets or when the price of the item in a purchase order is wrong. 

Conclusion 

APIs aren’t going away anytime soon. Business Central will always need OData and REST for rigid system-to-system heavy lifting to integrate between external systems

But for the new generation of intelligent automation: Co-pilots, agents, chatbots… the missing piece is the Model Context Protocol. It turns Business Central from a traditional ERP into a smart platform that any AI can talk to. 

For developers and architects: the message is clear: Start thinking about context. The end of the era of static integration is near. The era of intelligent agentic integration is upon us.